WeaknessASC-2026-0024Other weakness
Hidden instructions in invisible or disguised text
MediumMitigated
Unicode tag characters, zero-width characters, white-on-white text and HTML comments hide instructions that people cannot see but models read.
How to detect it
- Unicode tag or zero-width characters in content
- Text styled to be invisible, or comments in HTML and markdown
How to resolve it
- Normalize input and strip tag, zero-width and bidirectional control characters
- Extract visible text only from HTML before passing it to a model
How to investigate
- Decode the hidden text and find where the content came from
- Search other content from the same source for the same pattern
Sources
- Embrace The Red: ASCII Smuggler Tool: Crafting Invisible Text and Decoding Hidden Codes (January 2024)
- Håkon Måløy: Context Collapse, Part 3: AI Worming through Word (July 2026)
- OWASP Top 10 for LLM Applications 2025: LLM01 Prompt Injection
- OWASP Top 10 for Agentic Applications 2026: ASI01 Agent Goal Hijack
- MITRE ATLAS: AML.T0051 LLM Prompt Injection